[ntp:questions] How exact are the PPS pulses of common GPSses?

unruh unruh at invalid.ca
Thu Feb 14 17:06:41 UTC 2013


On 2013-02-14, David Taylor <david-taylor at blueyonder.co.uk.invalid> wrote:
> On 13/02/2013 09:00, Ralph Aichinger wrote:
>> After having my Raspberry Pi with PPS running, with an
>> offset and jitter value in the region of 1 to 5 microseconds,
>> I am wondering: How precise is the PPS signal of typical
>> consumer hardware (MTK chips, etc.)?
>>
>> Has anybody hooked up the PPS output of different consumer
>> GPSes to an oscilloscope and compared it against eachother
>> with a precision clock?
>>
>> Is the precision of the PPS pulse higher than what
>> the computer/ntpd daemon can usefully handle, or is
>> the PPS pulse the limiting factor?
>>
>> /ralph
>
> Ralph, I have compared a number of consumer-level timing and navigation 
> GPS/PPS devices and they are usually with 100 ns, and typically quite a 
> bit closer.  I haven't been able to compare against true UTC, but as 
> they are that close on several different manufacturer' devices I have no 
> reason to think that they are far from UTC.  At this level, BTW, the 
> circuitry controlling the rise-times of the PPS output matters when 
> comparing.
>
> What the time is /inside/ the Raspberry Pi is another question.  How 
> long does it take from PPS input to reading the counter?  I've seen 
> figures for the Raspberry Pi in excess of 10 microseconds.  So although 
> jitter may be reported as a few microseconds (or zero on one of my 
> Raspberry Pis), the virtual clock inside the PC, the time which NTP has, 
> may be ten microseconds or more away from "real" time.
>
> For my applications it doesn't matter, so it doesn't concern me, but 
> maybe it would concern others who may know more about this issue than I do.

To get an estimate you could output a pulse on one of RPi's GPIO pins as
soon as it gets the PPS and begins to process the input, and then compare them
on a high bandwidth digital scope. Certainly I would not be surprized at
1usec, but 10 would be surprizing. 




More information about the questions mailing list